Dr. Marco Zappatore

Topic: From Statistical Models to Generative AI: Exploring the Paradigm Shift in Machine Translation

Over the last decade, Machine translation (MT) has evolved profoundly, moving from rigid statistical systems to the breakthrough of neural machine translation and, more recently, to highly adaptive, Generative AI solutions powered by Large Language Models (LLMs), whose technical evolution is even faster. This represents a significant shift in the way human translators are supported by the technology and promises not only to redefine translation practices and quality but also to disclose new opportunities for language and translation education.

The capability of Generative AI to provide outputs that are more context-aware, nuanced, and creative than the previous generation of MT tools offers today more options to translators, students, and educators. Nevertheless, a variegated range of novel challenges has emerged, including hallucinations, biases, fluency-vs-accuracy trade-offs, and the redefinition of the role of human experts in AI-assisted translation.

During this talk, we will explore several technological enablers that can be employed in the translation landscape, such as interactive platforms for MT post-editing, AI-powered writing assistants, automatic tools for subtitling and dubbing, ad-hoc services for specialised translation (e.g., healthcare or legal), by considering their features and their suitability to specific application domains.

Finally, we will deal with the potential introduction of these aspects in translation teaching and training, with the aim of encouraging critical engagement of students with machine-generated outputs, as well as for providing learners with adequate skills for AI-augmented professional environments, where human expertise is still relevant and Generative AI tools should be perceived as a powerful ally and an amplifier of analytical skills, rather than a menacing replacement.